Employment, Wage Structure and the Economic Cycle: Differences between
5.5 Adjusting for Composition
Some of the differences in the labour market outcomes of immigrants and natives th at we have illustrated in the previous section could be explained by differ
ences in their composition. We now investigate this in more detail. We analyse whether, and to what extent, differences in outcomes over the business cycle can be explained by differences in skills, age structure, industry allocation or regional allocation. We do this by sequentially conditioning out differences between na
tives and the two groups of immigrants. In particular, we estimate the following model:
y i = x * a + £ Z ^ T>s + 4 g= natives, t —t\
O ECD, non—OECD
where y j is the outcome of interest for individual i belonging to group g (na
tives, OECD immigrants, non-OECD immigrants) in period f, Xft is a vector of additional controls like education, age, etc., and e^is an error term. Tt8 repre
sents the interaction of the group indicator g with year dummies for each year t. The parameters yf estimated for these interaction terms measure the aver
age outcome y for group g in period t, conditional on observables X 8. Simple re-parameterisation allows estimating the differences in outcomes over time rel
ative to a reference group. We estimate the following model by choosing as the reference group the native German and UK population, respectively:
ygit = X ?,< x+ £ £ ) W + £ y , 4 + <?f,
g=O ECD , t= t\ t= t\
non—OECD
dt are here year dummies for each year t. When restricting a to zero, the esti
mated parameters yf are the group mean labour market outcomes of OECD/non- OECD immigrants relative to the native population (picked up by //) as illustrated in the figures in the last section. By sequentially adding education and age, re
gional and, for Germany, industry controls, we eliminate differences in estimates of economic outcomes between our groups th at may be due to differences in these
observable characteristics. We plot the resulting estimates of yf in the figures that follow. This amounts to comparing immigrants and natives who are iden
tical in observables. In the initial estimations without controls, illustrated by the solid line, the only variable included in Xft (apart from a constant term) is an indicator for the gender of the individual. In the next step, represented by a dashed line, we add age, age squared and interactions of our education groups and year dummies. Finally, in the last step, illustrated by a dotted line, we also include interactions of region and year and, in the case of Germany, industry and year dummies. Unfortunately, the LFS data do not allow to condition on indus
try allocation since information on industry affiliation is not available for a large proportion of the unemployed - up to 40% of the observations in many years.
Notice th at we assume that all three groups respond in the same way to changes in the X f , so there are no group-specific cccoefficients (although we allow a to vary with time by using interactions of education, region and industry dummies with year dummies).
5.5.1 Unemployment
In the upper panel of Figure 5.4, we show the unemployment rates of OECD and non-OECD immigrants relative to the unemployment rates of natives for Germany. The solid line is the unconditional differential; the dashed and dotted lines control for differences in age and education structure, and differences in age, education, industry and regional allocation between immigrants and natives.
The figures suggest th at conditioning on age and education reduces the unem
ployment differential between Germans and immigrants in both groups; however, there remains a large difference and the cyclical pattern is clearly visible. Condi
tioning also on industry structure and regional allocation does not systematically change these differences except in the case of non-OECD immigrants during the period 1985-1995, where it tends to increase the unemployment differential and to some extent smooth the cyclical pattern. The figures that separate men and women look very similar to the pooled figure and can be found in Section 5.8.2 in the appendix to this chapter.
In the lower panel in Figure 5.4, we display the conditional and unconditional
Figure 5.4: Conditional unemployment rate differentials, Germany and UK
— age. education controls — age. education controls
..— a — age, education, region, industry controls -•••-a-....age. education, region, industry controls
IABS. p a r e n s agsd 2 i - n IABS. psrsons agsO a n
unemployment differentials for the UK. The differences between the conditional and unconditional patterns are much smaller than in Germany. This is not sur
prising, because the age and education structure of immigrants in the UK re
sembles those of the native population quite closely, as we have shown earlier.
Furthermore, although immigrants are highly concentrated in London, this is not an area with particularly untypical unemployment rates. Overall, we again see considerable differences between OECD and non-OECD immigrants, as well as the cyclical pattern in the early 1980s and 1990s which is particularly pronounced for the group of non-OECD immigrants.
5.5.2 Wages
In Figure 5.5, we display the unconditional and conditional log wage differen
tials for Germany and the UK. Again, the solid line depicts the unconditional differentials. As for the unemployment rate, we see a reduction in the wage
198
Figure 5.5: Conditional log wage differentials, Germany and UK
OECD vs Germany
-.1
I960 1965 1990 1995 2000
non-OECD vs Germany
- 2
2000
I960 1986 1990 1995
♦ no controls ♦ no controls
— » - ■ age. education controts — * —• age, education controls
age. education, region, industry controls ...a— age, education, region, industry controls
IAAS. ptfaons agad 2S-64 IASS, pwiont agM 2S-S4
OECD vs UK non-OECD vs UK
-.1
- .2
-I960 1985 1990 1995 2000 2005 1980 1985 1990 1995 2000 2005
■■■»— no controls — • — no controls
— age, education controls age, education controls
— a - - - age, education, region controls ... age, education, region controls
LFS. persona aged 26-54 LFS. persons egad 25-S4
differential between the two immigrant groups and natives for Germany when we condition on age and education, suggesting that part of the differential is due to differences in the age and education composition of the two populations. This is not surprising because we find large educational differences between groups in Table 5.2. However, there remain substantial differences, in particular for non-OECD immigrants. For this group, the differential decreases further when taking account of differences in industry and regional allocation. This suggests that non-OECD immigrants are particularly affected by the cycle not only be
cause of their low education level, but also because they have an unfavourable allocation across industries and regions. Conditioning on these, up to 1990, the wage differential vanishes entirely. However, after 1990, controlling for education, age, industry structure and regional allocation can only account for around 50%
of the widening wage gap between natives and non-OECD immigrants, thereby still leaving a gap of more than 10% unexplained in 2000. The gap between